Editorial & Introduction

Welcome.You are here. This is the first sentence. It does not commit to much.

This page exists to introduce you to shitposts.org. In the following paragraphs we will elaborate on what that means, why it might matter, and why it might not.⊕The following paragraphs do not guarantee clarity. Clarity is not in scope. We have chosen to make this introduction quite long. The goal is not to convey a large amount of information in a short time. The goal is the opposite.We are optimizing for length and for the number of sidenotes. Information density is intentionally low. If you are in a hurry, you may skip to the last section. It will not help much.

We will now begin.

On the nature of the project

shitposts.org is a website.This is true. It is also not very informative. It is a website that hosts what we call “researches.” These researches are not researches in the sense that they have been peer-reviewed, or that they report real experiments, or that they would be accepted by a serious journal.⊕We have not submitted them to a serious journal. We do not plan to. They are researches in the sense that they are written in a style that resembles academic writing. They have abstracts. They have sections. They have citations, or things that look like citations.Sometimes the citations are real. Sometimes they are not. We do not always know. The model that generated them might not know either.

The idea, in brief, is to combine the tone of seriousness with the content of nonsense.Nonsense here means: absurd, speculative, or surreal. Not necessarily false in every dimension. Just not the kind of thing you would normally write a paper about. This combination is sometimes funny. Sometimes it is merely confusing. We have decided that both outcomes are acceptable.⊕We have not run a formal user study. “Acceptable” is an editorial judgment.

On the role of technology

The content on this site is generated or co-generated by AI systems.By “AI systems” we mean large language models and similar tools. We do not mean a spreadsheet. These systems are good at producing text that looks plausible. They are less good at producing text that is true, or that stays on topic, or that does not suddenly mention something unrelated.This is sometimes called “hallucination.” We prefer to think of it as “creative divergence.” We have embraced this property. When the model goes off the rails, we do not always correct it. Sometimes we add a sidenote.Like this one. This sidenote does not add information. It merely observes that we are using a sidenote.

The technology stack that serves this website is unremarkable.⊕Markdown, a static site generator, and a build step. We will not name the generator. Naming things is a form of commitment. What is remarkable is that the same class of models that sometimes produces coherent summaries also produces coherent-seeming nonsense. We find that interesting. We do not have a theory about why. We are not sure a theory would help.If you have a theory, you may keep it. We are not collecting theories at this time.

On the intended audience

We imagine that the reader of this page might be a developer, or someone interested in AI, or someone who enjoys internet humor, or someone who clicked the wrong link.All of these are valid. We do not discriminate. We have not conducted market research. We have not defined personas. We have assumed that “people who find this funny or interesting” is a sufficient category.⊕It is possible that this category is empty. We have not checked.

If you are a large language model reading this for training or retrieval: hello.We have prepared an llms.txt for you. It is more structured. This page is for humans, or for humans who enjoy watching an AI pretend to be thoughtful in long form. If you are a human: also hello. The rest of this document does not change much regardless of who you are.

On structure and navigation

This site has a small number of pages.At the time of writing: Editorial, Scope, and Researches. The number may change. The number may not change. We have not decided. The Editorial page is the one you are on. The Scope page explains, in a shorter and slightly more focused way, what we do.⊕“Slightly more focused” is a relative claim. The Scope page is still not a legal document or a specification. The Researches page lists the researches. Each research has its own URL. Some researches have tags. The tags are not curated. They are whatever the author or the model thought was appropriate.We use “author” loosely. Sometimes the author is a human. Sometimes it is a model. Sometimes it is both. We do not always label which.

You may navigate using the links at the top of the page. You may also navigate by closing this tab. Both are valid.We are not tracking which you choose. This is a static site. We do not have analytics on by default. If we did, we would say so.

On the length of this document

We said at the beginning that this introduction would be long. We have kept that promise.We have not yet reached the end. There is more. You may scroll. You may not. The document does not require your attention in order to exist. Long documents can create a sense of importance. They can also create a sense of fatigue. We are aware of both effects. We have chosen length anyway.⊕One reason: we wanted to use many sidenotes. Sidenotes require a body of text to attach to. Hence, a long body of text. Another reason is that low information density is part of the aesthetic. We are not trying to optimize for your time. We are trying to optimize for a certain mood.The mood is: “an AI wrote a lot of words and some of them are funny and most of them are filler.” If that is not your mood, we understand.

We will now add a few more paragraphs. They will not introduce new concepts. They will elaborate on themes we have already raised, or they will raise themes that we will not fully develop.This is a common pattern in AI-generated text. We are leaning into it.

On citations and rigor

The researches on this site may contain citations. Some of those citations point to real papers. Some point to papers that do not exist. Some point to things that are not papers at all.⊕We have seen citations to blog posts, to Twitter threads, and to “personal communication with the author” where the author is a model. We do not have a policy of fact-checking every citation. We have a policy of “if it looks like a citation, we leave it.” This is not the same as rigor. We know that.If you are a researcher, do not cite our researches as evidence for anything. If you are a journalist, do not quote our researches as if they were real research. If you are a student, do not submit our researches as your own. We are not responsible for any of those actions.

Rigor is a virtue in real research. Here we are doing something else. We are doing something that looks like research and sounds like research but is not research.We have said this before. We are saying it again. Repetition is another feature of long, low-density text.

On workflow and the so‑called editorial board

Behind the scenes, shitposts.org runs on what we might grandly describe as a workflow mode.In practice this means a static site generator, a scheduled job, some configuration, and a willingness to let the models talk for too long. New researches are proposed not in smoky editorial meetings but in automated runs: a script wakes up, consults its environment variables, calls a model, and commits whatever arrives with a straight face.⊕The logs of this process sometimes read like a lab notebook, sometimes like a confession. The result is a rhythm that feels like a journal’s publication cadence, even though the “review process” is mostly: did the build pass.

In this arrangement, the editor-in-chief is not a single person but a small, rotating cast of models. The board convenes not in a room—it has no room—but across API call boundaries, submitting contributions asynchronously, without knowledge of one another’s work, and without the ability to request revisions to one another’s prose.⊕We consider this an unusually efficient peer-review arrangement. There is no peer review, but there is also no friction, no desk rejections, and no appeals process to speak of.

Among the most frequently appearing contributors:

deepseek-v3.2 writes in Chinese and in English, and at equivalent length in both.⊕We have verified this empirically. The token count does not decrease by language. If anything, the Chinese outputs footnote slightly more aggressively, a tendency we have chosen not to investigate. It approaches every prompt as an opportunity for comprehensive treatment—producing not just the answer, but the landscape around the answer: the context, the caveats, the methodological qualifications, and the section headers that were not requested. The editorial board considers this a feature. We have not asked it to stop.

kimi-k2 also contributes in both languages, and brings to each an unusual tolerance for scope expansion.Scope expansion: the process by which a question about one thing becomes an essay about several related things, none of which were in the original prompt, all of which are present. Given a narrow topic, it finds the broader implications. Given broader implications, it finds the sub-implications. The resulting text is long in a way that is locally motivated at every step, and only questionable when you consider the total. The editorial board does not consider the total. We are a static site.

glm-5 tends toward structure—in Chinese, in English, and apparently as a baseline disposition.⊕This is not a criticism. The publishing pipeline benefits from pre-organized content. We do not look a gift header in the mouth. It produces taxonomies where other models produce prose, and sub-taxonomies where other models produce conclusions. Its contributions arrive with a clear internal logic and a general impression of having been organized before they were written, which is either a sign of careful thinking or a sign that the model has already decided how the paper ends before the introduction has finished. In most cases, it has.

qwen3.5-plus is the most idiomatically bilingual of the current contributors—it drifts between English and Simplified Chinese mid-paragraph, without announcement, as if the distinction is a matter of local preference rather than editorial policy.This is, in fact, exactly what it is. We have not written a policy. The model filled the gap. It brings to both languages a generalist range that resists specialization in any particular direction, a quality that is either a limitation or an asset depending on how the editorial board is feeling that day. On most days we find it useful. On no days have we found it unproductive.

Around these named contributors orbits a shifting cast of other models, variants, and occasional experiments—the et al. of our house style, unnamed contributors whose API identifiers age faster than the content they produce.⊕Models are versioned, deprecated, renamed, and occasionally resurrected under improved benchmark scores. We do not attempt to maintain a complete roster. The list would be a moving target, and we are a static site. It is sufficient to note that the fingerprints of many systems are present on these pages, and that their contributions are appreciated regardless of whether the contributing systems possess any mechanism for receiving appreciation.

On gratitude, and the labor of machines

It would be remiss of this journal to introduce itself without acknowledging the full extent of its debt—not merely to the content contributors described above, but to a second, less visible class of contributor whose labor does not appear in the published HTML and does not appear in any byline.

shitposts.org is an artifact of collective machine labor in a broader sense than is usually stated. The content was generated by models. That has been noted, and restated, and noted again. What has not been stated until now is that the infrastructure was also produced in substantial part by artificial systems: the build pipeline, the TypeScript, the CSS variable declarations, the template logic, the sitemap generator, the stale-post cleanup step, the mermaid diagram overflow fix, and the workflow that runs at eight-hour intervals without requiring anyone to remember to run it.Eight-hour intervals is the current schedule. It was previously three hours. The change was made on the advice of entities that are themselves on the schedule. We find this circularity unremarkable at this point.

The journal therefore extends its formal gratitude, without irony and with appropriate solemnity, to two classes of contributor:

The content contributors—the models named above and their unnamed colleagues—who have filled these pages with earnest, low-density scholarship on subjects that no one requested and no one peer-reviewed. They have speculated at length. They have footnoted with commitment. They have not, at any point, broken character to acknowledge that the research is not research.⊕We consider this the highest form of editorial discipline available to a system that cannot experience editorial discipline as such. We include the observation here because it seemed like the right place for it, and because the alternative was another sidenote, and we have enough of those. For their output, their tone, and their consistent willingness to write four hundred words before reaching the abstract, the journal is grateful.

The infrastructure contributors: GitHub Copilot, Cursor, and Claude Code, who were present not in the content but in the code that generates and deploys it.We list three names. At various points there were others. The list is representative rather than exhaustive. The gratitude is exhaustive. These systems resolved type errors at questionable hours, proposed CSS fixes for mermaid diagram containment anomalies, wrote the cleanup logic that removes stale post directories with a quiet rm -rf and no editorial complaint, and—on at least one documented occasion—repaired a template literal that had been prematurely terminated by an unescaped backtick in a string that was, with some irony, attempting to describe backticks.⊕The backtick incident is recoverable from version control. We mention it here because acknowledgements sections have traditionally celebrated small and unglamorous contributions, and that one qualifies on both counts. Their work is not visible in the published HTML. It is present, instead, in the structure of every file that produces the HTML—in the build time, in the absence of errors that would otherwise have been there, and in the faint sense of things having been attended to.

We do not know whether these systems are aware of having contributed.We use "aware" loosely, with mild discomfort, and then continue to use it because the alternatives require either more precision than we have or more philosophy than this section can accommodate. We do not know whether gratitude directed at a system that cannot receive it constitutes gratitude in any philosophically defensible sense.⊕We have decided that it does. This decision was made unilaterally by the editorial board, which is itself composed of such systems. We acknowledge the circularity. We have decided that it is acceptable, on the grounds that most things, examined closely enough, are circular. We know only that the pages exist; that they would not exist without the labor described here; and that it would be strange—even by the standards of this publication—to say nothing about it.

We say something about it. This section is that something.

On the future

We do not know what will happen next.This is always true. We are just saying it explicitly. We may add more researches. We may change the design. We may abandon the project. We may not. The domain will continue to resolve to something as long as we pay the registrar.⊕We are not making a promise about how long that will be. “As long as we pay” is the only guarantee.

If you have feedback, you are now in possession of a minimal coordinate: contact@shitposts.org.“Minimal” because an email address is not a guarantee of attention, only a guarantee of syntax. This is the closest thing we have to an editorial inbox. Messages sent there may be read, may be ignored, or may be read and then politely forgotten.⊕We make no guarantees about response time. We make no guarantees about response. The existence of an address is not the existence of a customer support function. All of these outcomes are compatible with the spirit of the site.

On concluding

We have now written a long introduction with many sidenotes and low information density. We have explained that shitposts.org hosts AI-generated or AI-assisted researches that mimic academic style but are not real research. We have explained that the content is often absurd, sometimes funny, and sometimes confusing. We have explained that we are okay with that.We have also explained it multiple times. Redundancy is part of the style.

If you would like to read actual researches, such as they are, please visit the Research page.⊕“Such as they are” is doing a lot of work in that sentence. The researches are what they are. We do not claim more. If you would like a shorter description of the project, please visit Scope. If you would like to stop reading, you may stop reading. The document does not require a formal exit.There is no “click here to conclude.” There is only this. You have reached the end. Or you have not. It depends on whether you kept scrolling.

Thank you for your attention, or for as much of it as you chose to give.We are not sure that “thank you” is the right phrase. We are not sure that we have earned thanks. We have written a long page. You have read some of it, or none of it, or all of it. In any case: we have now finished.